On the Wages of
Temporary Help Service Workers in Germany
Michael Kvasnicka
¤, Axel Werwatz
yDepartment of Economics Humboldt University Berlin Spandauer Straße 1, 10178 Berlin
August 26, 2002
Abstract
This paper produces …rst econometric estimates for Germany of the contemporanous wage gap associated with Temporary Help Service (THS) employment, as well as the long-term e¤ects of THS work on the future earnings of workers. In addition, we present evidence showing that average male real earnings in the THS industry have declined by about 9% in the period 1975-1995. Our estimates of the overall and group-speci…c wage gaps attached to THS work are large, albeit signi…cantly smaller than those cited in the existing descriptive literature. Moreover, we …nd no evidence for any discernable negative average long-term earnings e¤ects of THS work. If anything, THS work seems to have aided workers in putting to a halt declines in their relative earnings observable before joining the THS market.
Keywords: temporary help services, wage di¤erential, long-term earnings JEL Classi…cation: J30, J31, J40
¤Corresponding author: Michael Kvasnicka, tel.: +49-30-2093-5667; e-mail: kvasnicka@wiwi.hu-berlin.de.
yThis paper has bene…ted from valuable comments at the annual meeting of the SFB 373 ´Quanti…cation and Simulation of Economic Processes´, a Brown-Bag seminar at Humboldt-University, as well as the IZA European Summer School in Labor Economics 2002. We are also grateful to the German Ministry of Labor and Social A¤airs, as well as the Institute of Employment Research (IAB), who have made the data for this project available. All remaining errors are our own.
1 Introduction
Growing at an average annual rate of more than 13% since 1973, temporary help service (THS) employment has been one of the fastest growing employment forms in Germany over the last three decades. Despite widespread economic stagnation and general labor market slackness, its pace of expansion even accelerated in the 1990s. Far from having tapped its full potential, forecasts (CIETT, 2000) predict undiminished growth of the THS industry in the …rst decade of the new millenium. Although the THS sector is still minute, accounting for just above 1% of dependent employment, it is far from being a negligible source of labor turnover and net employment growth in the economy. Roughly three quarters of a million of workers are engaged in the THS sector within a given year, more than twice its employment stock on a given day. In addition, more than one hundred and seventy thousand jobs have been created in the industry in 1995-2000 alone, a period in which employment that is subject to social insurance contribution fell by almost twice this …gure in the economy at large. In the year 2000, almost 6% of all newly formed employment relationships of the latter type were THS jobs.1 On a number of measures, therefore, THS work constitutes one of the most dynamic segments of the German labor market. This trend growth is discernable in most OECD countries. Since 1992, THS employment at least doubled in all member states of the European Union. At the same time, the industry has been deregulated lifting restrictive measures circumscribing its operation. In some European countries, in fact, this submarket has been legalized for the …rst time (Sweden in 1991, Spain in 1994, Italy in 1998, Norway in 1999).2 In Germany, the Hartz Commission set-up by the government to develop policy strategies to battle unemployment has put the creation of public THS …rms attached to regional employment o¢ces centerpiece in its recently publicized catalogue of recommendations.
However, concern has been raised about the quality of jobs generated in the industry, predomi- nantly about wages paid, but also about the long-term e¤ects of THS work experience on workers´
future employment and earnings prospects (ISA, 2001; German Parliament, 2000). Empirical evi- dence on relative remuneration levels in the THS sector in Germany hardly exists. Moreover, that which does exist is rather descriptive in nature than based on econometric analysis. Our knowledge of the long-term e¤ects of THS work on the future earnings of workers is even more limited. For Germany, evidence on this issue is non-existent. This analysis will aim to rectify these de…ciencies by providing empirical evidence on the earnings of THS workers, how they evolved over time, how they compare to wages paid outside this industry, and how subsequent long-term earnings of workers are a¤ected by a previous spell in THS employment.
1According to Houseman, Kalleberg, and Erickcek (2001), the growth in THS employment accounted for 10% of net employment growth in the US in the 1990s.
2For an overview on the European THS industry, as well as the national regulatory frameworks governing its operation, see, for example, Storrie (2002), CIETT (2000), or Bronstein (1991).
We proceed as follows. Section 2 introduces the THS market in greater detail and surveys arguments from economic theory that can be brought to bear on the issue of remuneration. Sec- tion 3 reviews the existing literature, and Section 4 describes the data. Section 5 and Section 6 contain the core of the empirical analysis. The former documents how THS wages have evolved as employment in the industry increased rapidly over the last decades. The latter contains an econometric analysis of average and group-speci…c wage gaps, as well as of the long-run e¤ects of THS work on the earnings of individuals. Section 7 discusses the main …ndings and concludes.
2 Theoretical Considerations
The THS market is ”a triangular arrangement in which a temporary work agency (TWA) hires a worker for the purpose of placing him or her at the disposal of a third party, the user enterprise, for a temporary assignment” (Bronstein, 1991). These ”transaction facilitators” (Smith, 1971) therefore act as labor market intermediaries between …nal labor supply and demand. In most countries, the THS …rm is the legal (de jure) employer of the THS worker3, and placements occur on the basis of a commercial contract between THS and client …rm that can be terminated both swiftly and at no cost to the client. This triangularity and the frequent change of work sites constitute the core features characterizing THS employment and set it apart from other forms of
‡exible labor such as employee leasing, …xed-term contract, casual, or on-call workers.
Proponents of this industry point to the bene…ts it bestowes on the …nal suppliers and deman- ders of labor, workers and client …rms. For the client, this submarket reinstitutionalises spot-market type (Ono, 2000), market-mediated fast access to labor services freed from otherwise applicable institutional constraints, such as collective bargaining, …ring costs, and the like. Client …rms there- fore bene…t from increased numerical ‡exibility in their sta¢ng policies. THS workers, in turn, are frequently recruited among the un- or nonemployed and are thus given access to paid work (entry-level jobs), otherwise often denied to them on the general labor market (CIETT, 2000).
Moreover, THS work can be a step-stone to regular employment (Houseman, Kalleberg, and Er- ickcek, 2001) by providing both work experience and multiple contacts with potential employers through successive placements. While workers can thus signal their ability to potential future em- ployers, clients can likewise screen THS workers during assignments for direct employment. The magnitudes of these ´access-to-work´ and ´step-stone´ functions of THS employment, however, are still insu¢ciently explored.4
3So, for example, in Germany, Spain, the Netherlands, France and to a lesser extent the UK, which together account for over 90% of THS workers assigned in the EU on a given day (CIETT, 2000). In the US, the legal situation is somewhat ambiguous on this count, but as Segal and Sullivan (1997) note, in the majority of cases the agency is in fact the legal employer.
4For Germany, Almus et al (2000) study the step-stone function of non-pro…t temping in Rhineland Paletine, and
…rst results for commercial THS …rms are presented in Schröder (1996). A paper investigating these two functions,
Opponents of this industry, on the other hand, stress the instability of employment relationships and the lack of career advancement opportunities in the THS industry. In fact, most employment relationships in the THS sector last less than a year. As can be seen in Figure 1, THS placements are markedly procyclical and display high seasonal ‡uctuations, with the latter in fact being only second in magnitude to those observed in agriculture (Storrie, 2002). Thus not only are current earnings of THS workers claimed to fall short of those obtained outside the industry, but future earnings of THS workers are seen to be adversely a¤ected too.
Figure 1: Monthly stock of THS Placements, 1973-1999
0 50000 100000 150000 200000 250000 300000 350000
73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99
Source: German Parliament, various years. Pre-1992, West Germany only.
Theoretically, however, it is far from clear that THS workers should in fact experience a contem- poraneous or future wage discount at all. According to the theory of compensating di¤erentials (see, for example, Rosen 1986), part of remuneration also comprises compensation for undesir- able job characteristics. If THS work entails a disproportionate accumulation of such negative attributes, workers should be compensated for them in the form of a relative wage markup. Em- ployment spells in the THS industry are in general of short duration, as pointed out above, and thus bear a greater risk of unemployment for the worker. This job instability, in turn, implies less intra-…rm career advancement opportunities and training. All of these features would demand a compensating wage di¤erential.
Lack of union coverage, high labor turnover, and the prevalence of general skills in the THS in- dustry, however, may prevent the formation of ´insiders´ that, endowed with su¢cient bargaininig
as well as, hitherto neglected, potential long-term e¤ects of THS work on employment paths of individuals is work in progress by one of the authors.
power, would ensure payment of these compensating di¤erentials. Moreover, if THS work ful…lls an ´access-to-work´ or ´step-stone´ function, then workers can be expected to implicitly ´pay´for these positive attributes in the form of a contemporaneous wage discount. As the fruits of this investment, i.e. improved future employment and wage prospects, accrue largely to the worker only, she and not the agency, according to human capital theory (Becker, 1964), is expected to bear the associated costs5.
The preceding discussion also entails implications for long-term earnings based on human capital theory. If THS employment indeed shares features of an investment for the worker, then long-term earnings of former THS workers should bene…t. If, on the other hand, insu¢cient on-the-job training or stigma e¤ects dominate, future earnings should be adversely a¤ected. In sum, a priori no clear-cut economic case can be made either about the likely sign of the wage di¤erential attached to THS work, nor about the implications of THS work experience for long-term wage growth. Many con‡icting arguments can be brought to bear on the issue, leaving undecided their likely net e¤ect.
Empirical analyis has to provide the answer.
3 Previous Research
The existing literature on relative wages in the THS sector is still sparse and inconclusive. Whereas the former de…ciency largely derives from the lack of adequate data for statistical analysis (IWG, 1995; Storrie, 2002), reasons for the latter are more multi-faceted, as the following literature review will show. Evidence on the e¤ect of THS work experience on the subsequent earnings of workers is even more limited. For Germany, it is lacking entirely. We will brie‡y review the main studies, their results and potential short-comings.
Rudolph and Schröder´s (1997) survey article on the German THS market, although only marginally touching the issue of relative compensation levels, constitutes the most valuable source of information for Germany on the subject to date. Using the employment statistics of the federal employment o¢ce (Bundesanstalt für Arbeit), the authors compare average earnings of THS and non-THS workers for the restricted sample of full-employed, male, West German workers, aged 25-35 for the years 1980, 1990 and 1995 and di¤erent occupations (high- and low quali…ed blue- and white-collar workers). They …nd that the overall wage gap associated with THS work has risen from -22.6% in 1980 to -36.6% in 1995. Moreover, the occupation-speci…c wage gap is larger for low-skilled than high-skilled workers. In a recent study (ISA, 2001) by the German Federal Union Association (DGB), new evidence by Rudolph is cited showing that this trend increase in the average wage gap has persisted throughout 1998. These …gures are also cited in the latest
5Although ´takeover-fees´ are sometimes charged by agencies to clients who want to permanently hire a THS worker, their payment is hardly enforceable.
quadrennial report of the German government on the industry (German Parliament, 2000). The magnitudes of these unconditional wage gaps are roughly in line with those reported in Storrie (2002) and those found in a survey conducted by the Institute of Economics and Society (IWG, 1995). Using the IABES 1975-1990, a one percent random sample of all workers registered by the social security system in Germany, Rudolph and Schröder (1997) also compare immediate pre- and post-THS earnings to those achieved while employed in the THS industry for all industry movers in 1980-1990. For 44.2% (73.6%) of workers, earnings improved when joining (leaving) the THS industry. Although these …gures are uninformative with respect to the long-term earnings e¤ects of having held a job in the THS industry, their magnitudes indicate that THS work experience might not necessarily lead to a deterioration of post-THS earnings at the individual level.
However, the German evidence cited above is merely descriptive in nature, referring to un- conditional mean di¤erences in earnings for all workers or broadly de…ned groups of workers.
THS employment, in fact, di¤ers systematically in its compositional structure from non-THS em- ployment. In Germany, THS workers frequently exhibit an accumulation of attributes commonly associated with lower earnings per se (see Table 2). Failure to control for the employment histo- ries, educational quali…cations, or occupations of workers, as well as potential ability di¤erences between THS and non-THS workers in a multivariate wage regression framework might therefore severely bias the estimates obtained.
That a more disaggregated analysis can lead to results quite di¤erent from those obtained through mere gross comparisons of average wage rates is evinced in a study by Cohen and Haber- feld (1993) who …nd occupation-speci…c wage di¤erentials of di¤ering sign and magnitude for the Israelian THS market. The authors run separate OLS wage regressions for four clerical occupations (typists, book-keepers, clerks and key-punch operators) in the early 1980s using two samples - one for THS workers drawn from a large THS …rm in Israel, the other for full-time, non-THS workers drawn from a 20% sample of the 1983 Israeli Census. Whereas the hourly wage rates of typists and book-keepers on average exceeded those of non-THS workers in the same occupations by 15%
and 9% respectively, those of clerks and key-punch operators in THS employment were 31% and 16% lower than the earnings of their non-THS counterparts. The average wage gap across all ocupations, however, amounted to 8%. Taken alone, the latter aggregate …gure clearly does not carry much information on pay structures in the THS industry covering up important di¤erences discernable at a more disaggregated level.
The importance to control for earnings-related attributes is also borne out in a study by Segal and Sullivan (1997) who use data on outgoing rotations of the Current Population Survey (CPS) from 1983-1993 to estimate the contemporaneous wage gap between THS and non-THS workers
in the US. In their preferred regression speci…cation, the authors run a panel wage regression with individual …xed e¤ects controlling for a host of earnings-related attributes like age, race, sex, education, union status, and one-digit occupation. The unconditional average wage gap for all collars can thus be reduced from -21.8% in a pooled OLS regression to a mere -3.1% in the …xed- e¤ect panel regression. Occupation-speci…c di¤erentials previously in the two-digits all fall below -6%. The estimated wage gap for white collar workers, though negative, is insigni…cant at the 10%
level. However, the data used su¤ers from severe drawbacks that call into question the reliability of their …ndings, as the authors themselves acknowledge in later work on the subject.6
In light of these data problems, the authors in a new study (Segal and Sullivan, 1998) take recourse to a di¤erent data set, a 10% sample of the quarterly wage records from the adminstrative
…les of the unemployment insurance system of the State of Washington covering the years 1984 to 1994. Compared to the survey-based CPS data used previously, this data set neither su¤ers from misreported industry a¢liation of THS workers, nor severe contamination of SIC 7363 with workers employed in PEOs (only about 3% in 1992). Lacking any demographic and occupational data, the authors use a regression framework frequently applied in the displacement literature (see, for example, Jacobson et al, 1993), controlling for both individual …xed e¤ects and time trends, as well as allowing for THS e¤ects in periods before and after the actual THS employment spell. The latter is accomplished by the inclusion of a series of dummy variables representing the number of quarters the current observation is apart from the THS employment spell.7 Unlike all other studies hitherto considered, this analysis therefore also examines the e¤ect of THS work on the subsequent (post-THS employment) earnings of workers. The authors …nd a THS wage gap of 10%, comparing THS wages to wages immediately before and after THS work or to the wages of non-THS jobs begun during the same period as workers were in the THS industry. Moreover, immediate pre- and post-THS relative earnings are of the same magnitude (ca. -5%) and the estimated wage gap becomes statistically insigni…cant after 5 quarters (base category: continuing non-THS job). Thus THS workers, on average, experienced a temporary dip in their earnings prior to joining the THS industry, but did not seem to be adversely a¤ected by their work experience in this sector. While the former …nding underscores the importance to control for the previous labor market histories of THS workers, the latter result indicates that neither inadequate training, possible stigma e¤ects
6THS workers frequently misreport their industry, leading to measurement error in the THS indicator. In addition, 25% of records could not be matched at all across years and were thus excluded from the analysis.
Identi…cation of THS work status is based on reported industry code SIC 7363 (”Help Supply Services”). But since 1987, SIC 7363 contains both THS and employee leasing …rms, also known as professional employer organization (PEOs). These workers are typically long-term employees of the …rms they are leased to and thus di¤er fundamentally from the group of THS workers. According to the 1992 Census of Services Industries about 23% of workers coded as working in SIC 7363 were in fact working in PEOs (Segal and Sullivan, 1998).
7These parameters, however, su¤er from an identi…cation problem due to the undi¤erentiated use of all available observations on workers over time in the regression analysis. For more details, and a solution to this problem, see Section 6.4.
attached to THS work, nor lack of …rm tenure, do seem adverse e¤ects of su¢cient quantitative importance to a¤ect the long-term earnings of workers. However, lack of typical human capital controls in the wage regressions make it impossible to ascertain, how the di¤erential varies across di¤erent groups (age, sex, education, etc.). The same missing data problem also cast a severe doubt on whether the problem of self-selection of workers into di¤erent (THS, non-THS) job types has been adequately dealt with. We will return to this issue of endogeneity in Section 6.1, where we discuss the identi…cation strategy employed for our empirical model.
Taken as a whole, a number of tentative conclusions can be derived from this literature survey that will form a starting point for our empiricial analysis. First, the gross of the available evidence indicates that average earnings in the THS sector are inferior to average earnings outside this sector. Second, di¤erences in the compositional structure of THS and non-THS employment seem to explain a large part of this wage gap. For Germany, however, regression estimates controlling for these di¤erences are lacking, so that current estimates of the wage gap based on descriptive analyses alone are likely to be severely biased. Third, evidence on the long-term e¤ects of THS work on the earnings of workers hardly exists. For Germany, in fact, it is simply non-existent. The available singular source (Segal and Sullivan, 1998) on this issue is based on US data whose results, however, might afterall shed little light on the functioning of the German THS market. As the legal framework governing the THS sector is far more restrictive in Germany than in the US, the usual quali…er beseting cross-country comparisons and transferability of results carries additional weight in the present context. We will adress both earnings dimensions in Section 6, where we provide empirical estimates of the contemporaneous wage gap attached to THS work, as well as the long-term e¤ects of THS work on the earnings of workers formerly employed in this sector.
Finally, surprisingly little attention has been devoted to how wages in the THS sector have evolved over time, as the industry expanded rapidly over the last three decades. This issue clearly deserves greater attention. We therefore start the empirical part of this paper in Section 5 with an analysis of relative and absolute wage trends in the THS industry.
4 The Data
The analysis is based on the IAB Employment Subsample (IABES) 1975-1995, a 1% random sample of all employees registered by the social security system in Germany over this period (see Bender et al. (2000) for a detailed description of the dataset). Employment information is based on statutory noti…cations by employers on their workforce to the institutions of the social security system. A noti…cation is required for the start and termination of an employment relationship, and hence any inter-establishment change. In addition, annual stock-taking noti…cations at the end of the year
update information on continuing employment. These statutory noti…cations are supplemented with additional information on the establishments issuing these noti…cations and unemployment periods that involve bene…t payments. Exact daily ‡ow information on the employment and unemployment history of more than half a million persons result in a total of nearly 8 million observations for the 21 years covered in the IABES. Large sample sizes are a desirable feature given the small share of the THS workforce in total employment. Moreover, typical problems beseting longitudinal surveys data, such as panel mortality due to non-responses, or memory gaps in restrospective questions are not encountered (Bender et al. 2000). The longitudial dataset contains a host of worker, job, and …rm attributes that make it well suited for the current analysis.
To exploit the longitudinal structure of the dataset for our statistical analysis, we construct an annual panel (sampling date: 30th of June) from this raw data to ensure that each and every sampling point in time corresponds to a unique noti…cation.8
THS Indicator: The IABES does not of itself allow identi…cation of an employment spell in the THS industry. For reasons of data con…dentiality, only two-digit industry identi…ers are included in the public use …le. A separate application for an identifying variable of THS employment spells (based on a restricted access to three-digit industry level information) was made and granted by the Federal Ministry of Labour and Social A¤airs and supplied by the Institute for Employment Research (IAB) of the Federal Employment Service. However, THS workers cannot be di¤erenti- ated from the permanent workforce of the THS …rm, as both are reported as working in WZW865
´Temporary Help Services´. But this short-coming is not too problematic, since permanent sta¤
only constitutes a small fraction of total employment in THS …rms (Rudolph and Schröder, 1997).9 Furthermore, only those …rms are listed as THS …rms in the IABES that do exclusively engage in this business.10
Measure of Earnings: The measure of earnings used in the IABES is average gross daily earnings per employment noti…cation. As the earnings information is based on noti…cations to the social security system, it is highly accurate. We express gross daily wages in prices of 1995 using an annual consumer price index. Although we only consider workers employed on the 30th of
8As information on continuing employment relationships is only updated with the annual stock-taking noti…cation at the end of the calendar year, changes in time-varying attributes like wages, occupation, and the like are only recorded for these employment relationships once a year. We have chosen the annual sampling date of 30.6. to make our …ndings time-consistent with o¢cial statistics (ANÜSTAT) on the THS sector in Germany, sampled on the 30.6. and 31.12.
9According to a recent survey of the THS industry in Germany (Lünendonk, 2002), the share of adminstrative personnel in the total workforce of the ten largest THS …rms in Germany amounted to less than 8% in 2001.
1 0In Germany …rms must obtain a licence in order run temporary help services. Firms can be of two types. Firms solely providing these services (´Hauptzweck Arbeitnehmerüberlassung´), or mixed enterpises that do also engage in normal production (´Mischbetriebe´). Only the former type are listed under WZW865. These employ about 85%
of the stock of THS workers (IW, 2000).
June in a given year, the wage information available are average daily gross earnings for the entire respective noti…cation period. As the latter may last up to a year, the use of an annual, instead of a quarterly price index is the more appropriate choice.11 Unfortunately, earnings information is both top-coded (censored), as well as truncated from below by design at the legal threshold for social security contributions. However, as less than 5% of observations per annum exhibit earnings that are top-coded, the former data defect is unlikely to be of major practical importance.12 We also ignore the latter restriction by con…ning our analysis to the population of immediate interest, i.e. workers in actual full-time employment.
Due to the sampling design, civil servants and the self-employed are not included in the IABES.
Omission of these two categories of workers, however, is not a severe shortcoming, as these are unlikely to constitute major alternative employment forms available to THS workers and thus a valid control group in our wage analysis. As the data set is based on obligatory noti…cations of employers about their workforce to the institutions of the social security system and the THS …rm is the legal employer of the THS worker, the dataset contains no information on client …rms.13 As a consequence, there is no information on the industry a THS worker is assigned to in a given placement. As industry a¢liation is regularly found to be an important determinant of wages, missing industry information for placements might lead to a bias in our analysis of the wage di¤erential. Although the issue cannot be de…nitely resolved, there are reasons to believe that such an omission is not too crucial for the validity of our estimates. THS wages are set independently of individual placements14, where the fee charged to the client …rm in a given placement generally takes the form of a mark-up over these wages(Storrie, 2002). Most of the variation due to di¤erent industry assignments is therefore likely to show up in fees charged rather than in wages received by the THS worker. Moreover, from a conceptual point of view, it is much harder to argue for a direct link between industry assignment and THS wages, as industry a¢liation of THS workers is two-dimensional (employment contract vs. work-site), one being time-constant, the other time- variant.
1 1We use the annual cost of living index for all private households in West Germany of the German Council of Economic Experts in their Annual Report 2001/2002.
1 2As a …rst robustness check, we ran our regressions both with and without those observations that are top-coded.
The results, in fact, hardly di¤er.
1 3Note that all existing studies reviewed in Section 3 su¤er from this drawback.
1 4Until recently, the durations of the employment contract of the worker with the THS …rm, which sets out compensation levels, and the …rst assignment of the THS worker to a client were not allowed to coincide by law, the so-called synchronization ban, aimed to achieve employment stability for workers in the THS industry.
Sample Selection: We have further restricted the sample to individuals that (1) are currently neither engaged in an apprenticeship, internship, nor do voluntary service, part-time or telework,15 (2) do not have any missings recorded in any of the variables used in the regression analysis, and (3) are working at establishments in West Germany, as the number of observations on THS workers in East Germany is too small for statistical analysis. In total, we can identify about …ve thousand THS workers. In addition, and to ease the computational burden, we have drawn a 1% random sample of non-THS workers. A list of the variables used in the regression analysis can be found in Appendix 1.
5 Wage Trends in the THS Sector
In a …rst step, we document how average earnings in the THS sector have evolved over time relative to average earnings in the labor market at large. For this purpose, we run the following simple regression
yit=¯0+ X20 s=1
¯st+ X21 s=1
°stDits +"it ,t= 1975; :::;1995. (1) Theyit denote the log of real daily gross earnings of workeri in year t. The ¯st are a set of year dummies that control for trend growth in wages, as well as cyclical patterns in aggregate wage dynamics. The Dits are annual di¤erential intercept dummies for THS work, so that their coe¢cients,°st, capture the potentially time-variant impact of THS work on wages in yeart. The
"it are error terms with the usual ideal properties. Except for the year dummies, this model is equivalent to a cross-sectional di¤erence in mean wages between THS workers and non-THS workers. The estimated annual percentage wage gap for full-time male THS workers in West- Germany for the years 1975-1995 is plotted in Figure 2.16
Covering two full business cycles, Figure 2 provides …rst, if yet preliminary information on the trend and cyclical properties of relative remuneration levels. First, for every year sampled, average male wages in the THS industry are inferior to average wages outside this sector. Moreover, the average male wage gap exhibits a marked upward trend over the 21 year time span, rising from just above 10% in 1975 to nearly 40% in 1995.
1 5In 1975-1995, workers in these positions account for only about 2% of THS employment, compared to 15% of employment outside the THS sector (see also Table A2 in Appendix 1).
1 6As the number of observations for female THS workers in the 1970s is quite small, we only report the result for male workers. The female wage gap, although throughout smaller than the male wage gap, displays the same trend and cyclical properties. However, its decline in the early 1980s seems to somewhat lead the cycle. As can be seen in Table 2, men constitute roughly four-…fths of all THS workers.
Figure 2: Average Wage Gap (%) for Male THS Workers, 1975-1995
0 5 10 15 20 25 30 35 40 45 50
1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 Year
Wage Gap (%)
Source: IABES 1975-1995. - - - denote 95% con…dence interval.
Second, a marked anticyclical pattern is discernable in the …rst, but not in in the second half of the time period under consideration. Output and the wage gap both peaked around 1980 and reached a trough in 1982, a pattern that is not discernable in the subsequent business cycle. The magnitude of the drop in the wage gap from 1981 to 1982 might also owe to the legislative changes coming into force in early 1982 that banned THS placements in the construction sector, which hitherto constituted a signi…cant fraction of total THS placements. Dragendorf, Heering and John (1988) attribute roughly one-third of the decline in the stock of THS employment over this period to this ban. Our own calculations based on the IABES indicate that average earnings of male THS workers holding jobs typically found in the construction sector were indeed about 19% lower than average earnings in the THS sector in 1981. The ban which came into force in early 1982 thus seems to account for a large fraction of the signi…cant, yet temporary reduction in the wage gap observed at the time.
To gain a more di¤erentiated view, we separately examine average real wage growth of men in THS and non-THS employment. Table 1 reports average annual growth rates in real earnings for both groups in 1975-1995 and in the sub-periods 1975-1984 and 1985-1995. Also included are the standard deviations of annual growth rates in these periods.
Table 1: Average Growth Rates p.a. in Male Gross Daily Real Earnings
Years 1975-1995 1975-1984 1985-1995
Mean growth rate (%):
THS -0.4 -0.5 -0.1
non-THS 1.5 1.5 1.5
Standard Deviation:
THS 4.2 1.8 1.0
non-THS 1.4 0.3 0.2
Source: IABES 1975-1995.
The …ndings reported in Table 1 are astonishing indeed. Whereas mean real wages of male non-THS workers have grown at an average annual rate of 1.5 %, the respective …gure for male THS workers is -0.4%. In other words, male THS workers have on average su¤ered a real loss in earnings of nearly 9 % over the 21 year period under consideration. The widening of the wage gap documented in Figure 2 therefore is driven by a signi…cant decline in average real earnings in the THS sector. This decline is more marked in the …rst, than the second subperiod, the former representing the early phase of growth and consolidation of the industry since its legal introduction in 1972.
Table 1 also provides information on the relative variability of wages, as measured by the standard deviation of annual growth rates. THS wages are far more ‡exible according to this measure in all periods considered. On the one hand, this …nding is to be expected. Contrary to the economy at large, THS wages are generally not collectively bargained. Nominal wage rigidities induced by annual wage settlements are therefore largely absent in this sector. Moreover, high turnover in the THS sector facilitates wage ‡exibility, as remuneration in new contracts can be more easily adjusted in accordance with changing economic conditions. On the other hand, however, this greater real wage ‡exibility is surprising, as THS employment is also far more volatile than non-THS employment. THS employment, as measured by workers currently in a placement, shows signi…cant year to year and seasonal changes often exceeding 30%. The THS sector therefore markedly outscores the general labor market onboth adjustment margins: the priceand quantity of labor.
One cannot but note the similarity of the German THS industry with the US labor market on this count. The latter, governend by largely unfettered employment at will and a low degree of unionization, produces both numerical ‡exibility in the sta¢ng policies of …rms and real wage
‡exibility. The di¤erence is that the THS industry in Germany constitutes a submarket imbedded in a general labor market beset with various institutional constraints circumscribing managerial prerogative in workforce matters. This comparative advantage of THS work might be a driving
force behind the rapid expansion of this industry, a hypothesis that has received empiricial support for the US(Autor, 2000). The link between this expansion and the real decline in average industry earnings documented above, however, is unclear, as demand of client …rms for THS services depends on the fee charged by the agency and not in any direct way on the wage paid by the latter to the worker.
A multitude of factors could drive this increasing wedge in remuneration levels. First, growth in the supply of labor to THS …rms might have outstripped growth in the labor demand of THS agencies, leading to excess supply of labor on the market which in turn caused wages to decline.
However, casual evidence suggests that agencies …nd it hard to recruit workers to meet the de- mand for their services and employment growth in the industry has been exorbitant. If anything, therefore, excess demand should have been the norm and competition among agencies for workers should have driven up wages in the industry. However, our time series evidence on wage trends does not conform with this prediction. Second, changes in the pro…le of job placements over time could have biased the compositional structure of THS employment towards lower-skilled workers.
If this specialization of the THS industry on the low wage sector occured against an opposing trend in the economy at large, we would naturally expect the unconditional wage gap to widen.
Third, the occupational wage distribution might have altered, adversely a¤ecting traditional skills provided in the THS industry. Fourth, competitive pressures among a sharply rising number of agencies in the expanding THS market might have led to severe price competition in fees charged, which in turn forced agencies to cut operating costs. As a labor intensive industry, this foremost implies reductions in the wage bill. With low barries to entry, sign…ciant growth in the number of agencies operating on the market, and continuing low levels of market concentration, such an explanation is quite plausible. To asses this hypothesis empirically would require time series data on fees charged, which, however, are not available. First evidence for the validity of this argument is provided by Barkume (1996) for the American THS market in the 1980s. Increased competition among agencies in this period of rapid employment growth in the industry has led to a decline in their gross margins, which in turn has spurred the demand for THS services by client …rms.
The study, however, does not comment about the implications of its …ndings for wages paid in the industry.
Descriptive evidence in support of the second hypothesis can be found in Table 2, which docu- ments the compositional structure and trend changes therein of both THS and non-THS employ- ment for various years. First, the THS workforce is on average younger and comprises relatively more male and foreign workers than the non-THS workforce.17 Second, the share of low- and high-
1 7Note that women and foreigners have been excluded from the regression, and therefore cannot account for the increasing wage di¤erential documented in Figure 2.
skilled blue collar workers is much lower in THS than in non-THS empoyment, and the share of white collar workers signi…cantly smaller. THS services thus remain geared to the industrial sector of the economy. Third, THS workers are more frequently recruited among the unemployed, even the long-term unemployed on both de…nitions used (>1/2 year or>1 year). Fourth, THS employ- ment exhibits relatively more workers of the lowest, and less of the highest educational/vocational attainment. These structural di¤erences apply to all years considered (except for recruitment among the previously long-term unemployed (> 1 year) for 1980, which is about zero for both groups).
Table 2: Compositional Structure of non-THS and THS Male Employment
Year 1980 1985 1990 1995
Sample Characteristics
(non-THS=1, THS=2) (1) (2) (1) (2) (1) (2) (1) (2)
Background Characteristics:
Average Age 39 31 39 30 38 32 39 34
Foreign (%) - whole sample 10 15 8 11 8 19 9 22
Women (%) - whole sample 33 27 33 23 33 23 33 21
Current Employment Characteristics (%):
low-skilled blue collar 24 33 24 31 24 36 23 36
skilled blue collar 42 57 42 62 41 52 39 52
high-skilled blue collar 4 0 3 0 3 0 3 1
white collar 31 10 30 7 31 12 35 12
Previous Employment History (%):
Unemployed 7 18 17 50 20 37 22 47
Unemployed>1/2 year 1 2 5 18 6 14 7 17
Unemployed>1 year 0 0 2 7 3 8 4 9
nonemployed>1/2 year 3 14 4 14 5 11 6 9
nonemployed>1 year (%) 2 7 3 10 4 7 5 7
Education & Vocational Training (VT) (%):
lower secondary degree, no VT 22 28 20 25 17 25 14 26
lower secondary degree, VT 74 69 75 72 76 69 76 70
higher secondary degree 0 2 0 1 1 2 1 0
higher secondary degree, VT 1 1 1 1 2 2 3 2
polytechnic degree 2 0 2 1 3 2 3 1
university degree 1 1 2 0 2 0 3 1
Source: IABES 1975-1995.
Moreover, in crucial respects these relative di¤erences have widened over time. The share of unskilled blue-collar workers in total employment shows opposing trends, while the increase in previously long-term unemployed workers is more marked for the THS sector. The decline in the share of the lowest educational attainment group in non-THS employment, on the other hand, is far more dramatic than that in THS employment. Thus both persistent strutural di¤erences in
employment composition and trends therein suggest a relative specialization of THS work on the low-skill sector. It remains to be determined, in how far these trend developments or the persistent structural di¤erences identi…ed account for the widening gap in remuneration levels documented in Figure 2. It could well be a mixture in both, in that the THS industry has increasingly recruited workers whose labor market prospects have relatively worsened over time. Wage ‡exibility on the THS market might have produced downward wage adjustment in light of excess supply of labor of this type.
6 The E¤ect of THS Work on Current and Long-Term Earn- ings
In this section we will go beyond the mere comparison of unconditional mean di¤erences in earn- ings between THS and non-THS workers. Three main questions of interest have been identi…ed in Section 3 that we now want to address. First, in how far do di¤erences in the compositional structure account for the signi…ciant unconditional earnings gap documented in the previous sec- tion. Second, how does the earnings di¤erential vary across di¤erent groups. And …nally, what long-term e¤ects, if any, does THS employment have on the future earnings of workers.
6.1 Identi…cation Strategy
The primary target of our empirical analysis is an estimate of the e¤ect of THS employment on the wages of THS workers. This ”treatment e¤ect on the treated” requires to construct an estimae of THS workers´ earnings had they not entered THS work. Estimating these counterfactual earnings is complicated by the fact that non-THS workers may not be an appropriate comparison group since, unlike in social experiments, individuals are not randomly assigned to THS work. Rather, workers self-select into THS employment based on observable or unobservable characteristics that may be correlated with their (potential) earnings in and out of THS work.
Heckman et. al. (1999) comprehensively discuss the assumptions and data requirements uner- lying the di¤erent approaches for constructing the counterfactual in this situation. Statistical matching, instrumental variables, switching regression and …xed e¤ects panel estimators are the most prominent among a wide range of methods. We employ the …xed e¤ects estimator through- out because it is well suited for our data and avoids some of the undesireable assumptions of its rivals. Unlike the matching appraoch, it allows for self-selection into THS work on the basis (of the time-invariant component) of the unobservables. It neither requires strong distributional as- sumptions, nor does it force the analyst to make the conceptually awkward distinction between covariates determining merely the selection into THS work, wages, or both. Moreover, while our
administrative data does not contain the wealth of demographic information needed to successfully apply matching algorithms, it does provide a fairly long panel of individuals´earnings and labor market status histories. The particular version of the linear panel data model employed below (and introduced by Jacobson et. al. (1993) in their analysis of the earnings losses of displaced workers) neatly summarizes this longitudinal information.
The ensuing analysis is con…ned to the period 1980-1990. Omitting the …rst years of the IABES should eliminate most of the downward bias introduced by left-censoring in duration variables em- ployed in the regression analysis. In addition, our preferred regression speci…cation for estimating both the conditional average wage gap as well as the long-term e¤ects on earnings of THS work (see equation 2 below) requires su¢cient pre- and post-THS observations (years) on individual work- ers for the indenti…cation of parameters in the model. For this reason, we also exclude the years 1991-1995 from the analysis. To further comparability of results obtained from various regression speci…cations, we have omitted these years entirely.18
6.2 Conditional Average Current Wage Gap
In the previous section, we have seen that signi…cant di¤erences exist in the compositional structure of employment between the THS industry and the general economy. Frequently these di¤erences entail an accumulation of attributes for THS workers commonly identi…ed with lower earnings per se. As documented in Table 2, the THS workforce comprises relatively more younger, foreign, low-skilled blue collar, less educated, and previously unemployed workers than employment in the economy at large. Comparing mere unconditional di¤erences in earnings therefore fails to disentangle the e¤ect of THS work status from that of attributes merely associated with THS work. The existing literature for Germany, as pointed out in Section 3, su¤ers from just this drawback.
To control for the latter, we run separate pooled regressions for men and women for the years 1980-1990 of the log of gross real daily earnings on a set of year dummies, a host of earnings- related attributes commonly included in wage regressions and a THS dummy variable.19 The estimated coe¢cient of the THS indicator therefore provides an estimate of theconditional mean di¤erence in earnings between THS and non-THS workers over the period 1980-1990, as we control for di¤erences in employment composition between these two groups. To control for unobserved heterogeneity among individuals, we, in addition, introduce individual …xed e¤ects (columns 2 and 4). The results are displayed in Table 3.
1 8See Section 6.4 for further discussion of this point.
1 9These earnings-related attributes comprise types of ocupation (four groups), measures of educational and vo- cational degress (six categories), two age categories, current job tenure, establishment size (7 categories), previous labor market status, and previous employment, unemployment, and non-employment durations. See also Table A1 in Appendix 1.
The unconditional mean real earnings gap for men and women in this period amounts to 30% and 25%, respectively. As Table 3 shows, controlling for di¤erences in the compositional structure reduces these estimates of the wage gap to about 20% for both sexes. One third (one fourth) of the unconditional wage gap for male (female) THS workers can be thus be accounted for. Controlling in addition for permanent unobserved individual heterogeneity by running a …xed e¤ect panel regression reduces the estimated wage gap to under 17% for men. For women, however, the estimate obtained is not materially altered. As expected, therefore, signi…cant reductions have been achieved in the estimated wage gap, once di¤erences in employment composition have been adequately controlled for. The estimate for men is halfed, the one for women reduced by one fourth. Compared to the existing descriptive evidence for Germany reviewed in Section 3, our regression estimates of the wage gap attached to THS work are signi…cantly smaller, albeit still large.
Table 3: Conditional Mean Real Earnings Gap, 1980-1990 Dep. Variable: log real daily gross earnings
Estimated coe¢cients:
Men Women
OLS FE OLS FE
THS contract ¡:227¤¤¤
(0:04) ¡:184¤¤¤
(0:04) ¡:213¤¤¤
(0:008) ¡:215¤¤¤
(0:007)
Hausman testa 1239:22¤¤¤ 429:28¤¤¤
R2 0.492 0.308 0.414 0.318
N of cases 39240 39240 15843 15843
Source: IABES. Years 1980-1990. *** indicates signi…cance at the 1% level.
a: Random e¤ects versus …xed e¤ects regression.
6.3 Conditional Group-Speci…c Current Average Wage Gap
So far, we have only estimated the average e¤ect of THS work on current wages. Now, we explore how the wage gap varies across di¤erent occupations, skill categories, and previous employment status. For this purpose, we include interactions of the THS dummy with these categories into our …xed e¤ect panel model. As before, additional covariates are included in the regression, but omitted from the regression output in Table 4 (see footnote 19 above).
Male low-skilled blue collar THS workers experience an earnings discount three times larger than male white-collar THS workers, and almost twice as large as that of male skilled blue-collar THS workers. For female THS workers, these di¤erences are far less marked. In addition, for skilled blue and white collar female THS workers, the respective discounts are virtually identical.
For non-THS workers, both types of blue collar workers earn only about 11% less than white collar
workers. Occupational wage dispersion among THS workers is thus signi…cantly more pronounced.
The return to …rm tenure for men in THS employment exceeds that in non-THS employment by half a percentage point. Although the estimate is statistically highly signi…cant, economically it is not. Several years of THS employment would be required to but half the wage gap. Moreover, only rarely do THS workers reach such tenures. Previous unemployment duration in years has less of a negative e¤ect on the current earnings for male - but not female - THS workers than for non-THS workers. For male non-THS workers, the return to an additional year spent in unemployment is - 3% (omitted from the regression output in Table 4).
Table 4: Group-Specific Mean Earnings Gap Dep. Variable: log real daily gross earnings
Estimated coe¢cients:
Men Women
THS£unskilled blue collar ¡:285¤¤¤
(0:006) ¡:267¤¤¤
(0:015)
THS£skilled blue collar ¡:164¤¤¤
(0:005) ¡:224¤¤¤
(0:062)
THS£white collar ¡:093¤¤¤
(0:011) ¡:222¤¤¤
(0:009)
THS£firm tenure :006¤¤¤
(0:002) :014¤¤¤
(0:004)
THS£previous unemployment duration :024¤¤¤
(0:006) :014
(0:013)
THS£previously nonemployed>1yr. ¡:006
(0:015)
:022
(0:026)
R2 0:295 0:306
N of observations 38386 15812
N of persons 6116 2767
Source: IABES. Years 1980-1990. Base category: white collar non-THS workers with lower secondary degree and no VET who were employed prior to their current job and are aged 18-29.
*** indicates signi…cance at the 1% level.
6.4 THS Work and Individual Earning Paths
In our preferred speci…cation, we explicitly allow THS employment to have lasting e¤ects in that we include a set of dummy variables T HSk;it with k equal to the number of years the current observation is apart from the THS employment spell. Thus k>0 (k<0) identi…es observations of workers k years after (before) their THS employment spell (k = 0). We include leading indicators for THS work, i.e. T HSk;it for k<0, as THS workers might have experienced a deterioration in their earnings prior to entering the THS market, caused, for instance, by job loss, plant closure, or general career setback. Part of the wage gap observed for THS workers, therefore, might be attributable to these events, rather than THS work per se. Although we have controlled for the
previous labor market status and history of THS workers in the previous regressions, we have not explicitly di¤erentiated between pre-THS earnings far removed from the THS employment spell and those immediately preceding it. If THS workers do indeed su¤er such a dip in their earnings before taking a job in the THS sector, then prior earnings of the individual worker far removed from the actual THS work spell do not constitute a valid benchmark in the calculation of the THS wage gap. More recent earnings rather than those of the distant past are likely to re‡ect actual current earnings potential and thus constitute a more appropriate comparison wage. Our regression model thus takes the form
yit=µt+ Xm
k=¡m
°kT HSk;it+xit¶¯+®i+uit ,t= 1980; :::;1990 (2) whereµtis a set of year dummies and we eliminated from our sample workers with more than one spell of THS work20. The vectorxitcontains earnings-related attributes, listed in footnote 19, that were also used in the previous regressions. The above model is identi…ed by the assumption that
°k= 0injkj> m, i.e. in periods more than k years removed from the THS work spell. In addition, identi…cation of the parameters°k rests on the correct speci…cations of the indicatorsT HSk;itfor each individual included in the regression. This requires su¢cient observations on individuals´ THS status, both before and after those observations actually included in the regression. For instance, a worker might be in THS employment for the …rst time in 1996. However, as the IABES only covers the period 1975-1995, this worker would exhibit T HSk;it = 0for all k<0 and t smaller or equal to 1995, and thus be misclassi…ed in the analysis as a no-time THS worker. Omitting the last years of the IABES from the actual regression, but using their information on THS membership for creating our leading indicators for future THS work for the years included in the analysis, avoids this misspeci…cation. Segal and Sullivan (1997) in their study on the THS market in Washington State have failed to control for this problem in their speci…cations of both leading and lagging indicators for THS work. As a consequence, four out of the eleven years of the data used in their analysis are potentially a¤ected by this misspeci…cation.
Variants of this approach has been widely used in the displacement literature (see, for example, Jacobson, LaLonde, Sullivan, 1993). However, a notable di¤erence exists in the use of the estimated lead coe¢cients in the calculation of the wage penalty associated with job loss in the displacement literature and the wage gap in our application. Displacement or …rm closure might already have e¤ects on the earnings of workers prior to the date they are ultimately laid o¤. Earnings in periods immediately preceding displacement might therefore not be independent of the ensuing
2 0Almost 70% of workers who at some point in time work in the THS sector exhibit only one spell in this industry, as identi…ed by our annual industry identifer.
displacement event and therefore do not qualify as a valid benchmark in the calculation of total losses incurred by job loss. If so, then earnings observations su¢ciently far removed from and thus una¤ected by the actual displacement event are to be preferred.
In our application, however, THS work status cannot be assumed to a¤ect the earnings of workers prior to their engagement in the THS industry, but only their current, and possibly, future earnings. As pointed out above, therefore, the most recent earnings observation prior to the THS spell are more likely to re‡ect the labor market (earnings) opportunities of THS workers, and are thus to be prefered as the appropriate benchmark. The estimated coe¢cients of the leading and lagging indicators for THS spells along with the one of the actual spell are plotted in Figure 3.
The regression output is included in Appendix 2. As can be seen in the …gure, setting jmj to 5 periods is not restrictive, as the estimated coe¢cients are statistically insigni…cantly di¤erent from zero at times far removed from the actual spell in THS employment.
Figure 3: Relative Pre- and Post-THS Real Earnings of Men
-0.3 -0.25 -0.2 -0.15 -0.1 -0.05 0 0.05 0.1
-5 -4 -3 -2 -1 0 1 2 3 4 5
No . o f yrs. o bse rv ation is apart from THS sp e ll
wage differential % (coeff.)
Source: IABES 1975-1995. Base category: Continuing non-THS jobs of workers never in THS employment in 1975-1995 and THS workers>5 years apart form their actual THS spell.
- - - denote 95% con…dence interval.
Average earnings of male workers start to fall well in advance of the actual THS spell, drop signi…cantly at the time workers hold a THS job, but recover fully and immediately to average
earnings of the base category thereafter. While the same earnings pro…le applies to women, both the magnitude of the decline in their pre-THS earnings and the actual THS wage gap are more pronounced. This earnings pro…le is also consistent with that estimated by Segal and Sullivan (1997) for THS workers in the state of Washington. However, the magnitude of their estimates of both pre-THS earnings in k = -1 and the THS wage gap is merely half that reported in Figure 3.
Moreover, relative pre-THS earnings start to decline but in the last year before the actual THS spell, whereas we …nd the decline to set in several years in advance. While the reasons for this are unclear, it makes the subsequent recovery for German THS workers all the more remarkable, both in terms of the magnitude involved and in light of the duration of pre-THS earnings decline.
The observed earnings pro…le is surprisingly robust across di¤erent subgroups of THS workers.
Appendix 3 contains graphical representations of the estimated pro…les for di¤erent age cohorts, and vocational quali…cations.
We have pointed out that previous earnings of THS workers in the more distant past are unlikely to be a valid benchmark for the calculation of the contemporaneous wage gap attached to THS work, as these do not capture the earnings potential of the individual at the time he holds a job in the THS sector. As the theoretically preferable counterfactual earnings in k = 0 itself are unobserved, we take the most recent pre-THS earnings as our best estimate instead. The estimated wage gap thus arrived at amounts to 12% for men, and 10% for women.21 These estimates are smaller than those reported in Table 3, and signi…cantly smaller than those cited in the existing literature on Germany. In addition, as pre-THS earnings steadily declined up to the actual THS employment spell and might have continued to do so, our estimates have to be seen as an upper bound for the true average wage gap attached to THS work. Estimates of the contemporaneous wage gaps for various sub-groups of workers are contained in Table 5 below. A number of …ndings emerge. First, women joining THS work earn signi…cantly less than women in non-THS work prior to their actual engagement in the industry. As a direct consequence, and although their relative earnings in THS work are markedly lower than those of their male counterparts, the estimated wage gap for women in THS work is smaller than that for men. Second, younger workers su¤er more of a wage discount than older workers from holding a THS job. This could be driven by lack of su¢cient work experience for the former group, arguably necessary for the immediate productive placement of THS workers. And …nally, workers without any vocational degree su¤er a signi…cantly larger wage gap than those who do have completed such training, a …nding that is consistent with human captial theory.
2 1Calculated, respectively, as 100*(exp(-.217-(-.088))-1) and 100*(exp(-261-(-.152))-1) using the estimated regres- sion coe¢cients of Table A3.
Table 5: Group-Speci…c Mean Real Earnings Gap for THS Workers, 1980-1990 Workers °^¡1 ^°0 °^0¡^°¡1 Wage Gap (%)
Men ¡:088¤¤¤
(0:011) ¡:217¤¤¤
(0:098) ¡:129¤¤¤
(0:011) ¡12:11
Women ¡:152¤¤¤
(0:022) ¡:261¤¤¤
(0:016) ¡:109¤¤¤
(0:022) ¡10:31
born in 1955-1962 ¡:023
(0:020) ¡:214¤¤¤
(0:014) ¡:191¤¤¤
(0:021) ¡17:41
born in 1930-1954 ¡:107¤¤¤
(0:013) ¡:247¤¤¤
(0:010) ¡:140¤¤¤
(0:015) ¡13:09
with VT ¡:092¤¤¤
(0:013) ¡:184¤¤¤
(0:010) ¡:092¤¤¤
(0:014) ¡8:77
w/o VT ¡:082¤¤¤
(0:020) ¡:283¤¤¤
(0:016) ¡:201¤¤¤
(0:021) ¡15:21
Source: IABES. Years 1980-1990.
Last four regressions for men only. *** indicates signi…cance at the 1% level.
This deterioration in pre-THS earnings of workers could well have in‡uenced their decision to work in the THS sector, even if this, on average, temporarily implied a large scale earnings loss. One of the arguments for an expected negative earnings di¤erential in THS work we presented in Section 2 has been the claim that THS jobs could act as a stepping-stone to regular jobs, aiding individuals in their job search and improving the match quality of subsequent employment relationships thus formed. While THS workers implicitly pay for these services in the form of a contemporaneus wage discount, on average they seem to recoup their outlay on this investment through improved subsequent earnings. This hypothesis, while not the only one, is consistent with the evidence in Figure 3. THS workers seem indeed on average able to permanently stop the downward trend in their pre-THS earnings pro…les. This would attest the THS industry an important intermediary function in the general labor market, i.e that of a successful employment agencies geared towards the …rst labor market. If this is indeed the case, then the documented wage gap attached to THS work, rather than a cause of concern for social policy makers, constitutes a valuable investment of the worker in her future labor market prospects.
While consistent with our …ndings, it may well be that other factors account for part of the contemporaneous wage gap attached to THS work. Collective wage bargaining in the THS sector, unlike on the general labor market, is the exception rather than the rule. Wage ‡oors introduced by collective wage determination are thus largely absent on this submarket. The observed wage gap, therefore, rather than being caused by THS work per se, might in part re‡ect the true market wage as determined by competitive market forces for the type of labor typically provided by THS agencies. Such an explanation is consistent with the ´access-to-work´ function of THS …rms, where agencies o¤er low-productivity workers the opportunity to price themselves back into work (entry-level jobs), access to which might be denied to them on the general labor market for lack
of su¢cient wage dispersion.
Our results in Figure 3 do not allow to adjudicate between these two hypotheses. However, the observable negative trend in earnings prior to THS work is put to a halt in subsequent employment.
Permanent earning losses of THS workers, due to stigma e¤ects or low skill upgrading caused by insu¢cient on-the-job training in the industry are thus on average not discernable.
7 Conclusion
We have investigated the contemporary wage di¤erential associated with THS employment, as well as the long-term earning paths of former THS workers. Average wages of men in the THS industry have fallen in real terms over the 21 year time span under consideration, leading to an overall real loss of nearly 9%. The average male THS worker in 1995 thus earns less in real terms than the average THS worker in 1975. This decline is all the more remarkable given that it has occured against the background of strong employment growth in this sector. We have but tentatively explored potential reasons behind this trend change in earnings. Changes in the compositional structure of THS employment, worsening labor market prospects of those type of workers traditionally employed in the THS sector, or …erce price competition between increasing numbers of agencies have been identi…ed as potential casues. Future research will have to adjudicate between these competing, yet potentially complementary explanations.
We have seen that a large part of the unconditional average wage gap associated with THS work can be explained by di¤erences in the compositional structure of THS and non-THS employment.
Controlling in addition for individual …xed e¤ects reduces the male, but not the female THS wage di¤erential.
Moreover, the wage gap has been found to vary signi…cantly across di¤erent occupations, with low-skilled blue collar workers su¤ering the greatest discount. It has been argued that such a discount might be the price to be paid for ´access to work´ otherwise denied to these low-skilled workers on the general labor market. As wages are not collectively bargained in the THS industry, they are free to adjust to imbalances in the supply and demand of types of labor.
In light of the short durations of THS employment spells, however, it is the long-term e¤ects of THS work on the earnings of workers, and not the wage gap attached to THS work that is of primary interest. THS workers have been found to su¤er a marked deterioration in their earnings prior to holding a job in the THS industry, yet do not seem to be adversely a¤ected by the latter in their future earnings prospects. Quite to the contrary, with the THS employment spell this earnings decline, on average, has been all but eliminated. Moreover, setting THS earnings in relation to the latest pre-THS earnings of workers as a proxy for their counterfactual earnings potential at the
time they hold a job in the industry substantially reduces our estimate of the wage gap attached to THS work to 12% for men and 10% for women. It has to be noted that part of these wage gaps might be attributable to the need of the agencies to recoup …xed outlays on recruitment, while being faced with exceedingly high labor turnover in the industry. Short employment durations preclude the possibility to spread these costs across time, and might hence necessitate larger per- period wage discounts. In adition, idiosyncracies of the legal framework governing the operation of the German THS market, such as the re-employment ban of recently …red THS workers, are likely to increase hiring costs.
While this analysis has provided improved quantitative estimates on one dimension of earnings (THS wage gap) and derived …rst evidence on another (long-term earnings e¤ects of THS work) for Germany, the driving factors behind these …ndings are still in the dark. Future research will have to investigate these in greater detail and try to adjudicate between the rival explanations hinted at in the text. Until then, whether or not a wage gap for THS workers gives reason for concern, let alone justi…es the use of corrective policy measures, is a question begging the facts.
